duckdb
Why remove the prepared statements from cache?
Maybe I am misunderstanding the logic here but I wonder why duckdb removes the statement from the cache instead of just getting it? I would expect a call to https://paritytech.github.io/try-runtime-cli/hashlink/lru_cache/struct.LruCache.html#method.get.
https://github.com/duckdb/duckdb-rs/blame/2bd811e7b1b7398c4f461de4de263e629572dc90/crates/duckdb/src/cache.rs#L142
I didn't write this code but from https://github.com/duckdb/duckdb-rs/commit/80f6f90c129a8ce882420680064b7e62f938122d it looks like the idea is that the statement is mutable while it's in use, perhaps at the DuckDB level. When you get the statement from the cache, the caller gets effectively exclusive use of it. But when the CachedStatement is dropped, it automatically puts itself back into the cache.
It seems a little strange compared to e.g. cloning when things are taken from the cache, or just keeping only immutable representations in the cache. But probably the C API constrains how things can be done.
The use of remove() instead of get() is intentional and necessary. Prepared statements are mutable during execution - RawStatement::execute() takes &mut self and modifies internal state. If we used get(), the cache would maintain a reference to the statement while also giving out a mutable reference to the caller, violating Rust's borrowing rules.
This is the same pattern used by rusqlite (which this crate was modeled after): https://github.com/rusqlite/rusqlite/blob/9d701b626fd15ab6c712ed2b082bf36a5b48bba3/src/cache.rs#L143
